As the aquaculture industry is growing, more sophisticated technology is required to monitor farms and ensure sustainability and good fish welfare. Using behaviour as a non-invasive form of monitoring, in combination with artificial intelligence and machine learning, can allow for higher control over farm management. The goal of this study was to identify changes to farmed Atlantic salmon ( Salmo salar ) behaviour related to fish health status. For this, cameras were deployed within all cages in each of 2 Atlantic salmon marine farms located in the Outer Hebrides, Scotland. One ‘study’ cage in each farm was equipped with 5 and 4 cameras, for sites A and B , respectively, for higher spatial coverage of fish behaviour throughout the cage. An algorithm was created by Observe Technologies to process video footage from these cameras and transform it into behavioural data, termed ‘activity’, which encompasses fish abundance, speed, and shoal cohesion. Daily validation occurred for the duration of the study, whereby experts compare the videos to output from the algorithm. Additionally, gill health scores were evaluated weekly at both sites through the visual sampling of 10 fish in each cage (0 = healthy gills, 5 = necrotic tissue). During summer 2023, gill health issues arose at both farms, leading to fish stress which was evident in the behavioural data. Prior to the rise in gill health scores, the average (± standard deviation) fish activity levels were 36.8 ± 11.3 % and 32.5 ± 10.5 % for each study cage in sites A and B , respectively. Following the increase in gill health scores , the fish activity rose significantly with a mean of 57.5 ± 14.6 % and 51.1 ± 19.0 %, respectively (U = 2 106 , p < 0.001 and U = 5.9 105, p < 0.001 ). Consequently, there was a significant correlation between PGD scores and activity at both farms (p < 0.001; R2 = 0.42 and 0.62, respectively; Fig. 1). The observed increase in fish activity was observed in all cages at each farm and corresponded with a shift towards the centre of the cage, indicating shoaling behaviou r commonly associated with a stress response. Subsequent investigations will explore the feasibility of an early-warning indicator for compromised fish health, potentially preceding visual detection by farmers.